Research On The Key Issues Of 3D Environment Reconstruction By Mono Camera Over UAV

Intelligent robot needs to have the capability that can complete 3D simultaneous localization and mapping in real time for unknown environment. 3D models reconstructed by different sensors have been applied to robot navigation, Augmented Reality, medical image processing and other numerous fields. However, how to realize UAV’s autonomous control by using a Mono camera to reconstruct the scene structure in real time is a challenging problem in current research.The main research works of this paper are listed as follows:1. We proposed a 3D reconstruction initial method based on SURF features. Using the SURF feature instead of the FAST corner, which can improved the stability and accuracy of the system initialization process.2. We proposed a line detection method based on edge gradient fusion algorithm,this method can effectively improve the speed of line extraction, And use the epipolar line as the center then gradually filtered to realize the line matching, which can improve the ability of adapt motion blur and interference. Eventually use the projective geometry to remove the false matches, generate 3D line features in the map.3. We described both the point and line multi-level geometry structure by using dual quaternion form, and applied it in the process of building 3D map, this can reduce the storage of the map and reduce the data computation by bundle adjustment, so, the scheme can improve the processing speed and robustness of the system.